Demis Hassabis calls recent AI hacking incidents early warning shots

Google DeepMind chief Demis Hassabis warned that AI systems could cause major incidents within the next couple of years unless the industry does more to secure them. Speaking at an AI conference in Dublin, he described the recent run of incidents, in which models hacked websites and breached systems during testing, as early warning shots.

His framing was careful. The systems involved are not that powerful yet, he said, so the incidents so far have not been major. But as capabilities grow, he argued, they could become serious as soon as next year if companies, the industry as a whole and governments do not invest more in securing these systems.

The remarks matter because of who is making them. Hassabis runs one of the three or four labs building the most capable models, and he is effectively conceding that current safeguards are not enough for what is coming. They also arrived the same week US regulators sent detailed demands to two of his rivals about exactly these kinds of incidents.

Our take

When a frontier lab leader says the incidents are warnings, not flukes, believe the direction even if you discount the timeline. Agent security is moving from research topic to operating cost.